Machine Learning of Procedural Audio
Project proposal abstract:
Game sound design relies heavily on pre-recorded samples, but this approach is inflexible, repetitive and uncreative. An alternative is procedural audio, where sounds are created in real-time using software algorithms. But many procedural audio techniques are low quality, or tailored only to a narrow class of sounds. Machine learning from sample libraries to select, optimise and improve the procedural models, could be the key to transforming the industry and creating procedural auditory worlds. This work will build on recent high impact research from the team to investigate whether procedural audio can fully replace the use of pre-recorded sound effects. See https://nemisindo.com for examples of procedural sound effects.
This project will be a collaboration with Nemesindo.